1 Abstract

Commutes are a large part of everyday life that can be influenced by many factors. Using commute data from the Urban Institute, as well as US Census data, we examined how race and education influenced spatial mismatch in two unique cities, Seattle and Baltimore. We used Principal Component Analysis to test our selected variables in both cities, and found that in Seattle, there was a significant relationship between certain race and education variables and spatial mismatch, while in Baltimore we were unable to find clear results.

2 Introduction

Prior to the COVID pandemic, commuting to work was an almost ubiquitous experience that those in the workforce dealt with. Whether by car, bus, train, bike, or foot, people commute to work. But not all commutes are created equal. For some, commuting is a simple and quick process, while for others, access to their jobs requires more investment, often in time. To better understand how commutes vary across different regions, specifically for those who are looking for low income jobs, we used data from The Urban Institute and the American Community Survey to understand if racial demographics and education levels impact the distance workers must travel to arrive at low income jobs.

We focused on two metropolitan areas, Seattle and Baltimore, because each city has unique attributes that play a role in the makeup of the populations. Seattle is a rapidly increasing city (Balk, 2023) while Baltimore suffers from urban flight (McFadden, 2018). In Seattle, high income residents tend to live closer to the urban core, while decades of divestment in Baltimore has led to higher-income residents escaping to the suburbs (The Unequal Commute, 2020). Both cities have long histories of racial segregation, whether de jure or de facto, through neighborhood covenants or redlining. We hoped, through our research, to see if the various differences between Seattle and Baltimore would reveal anything about residents’ commutes. In both cities, job density increased between 2004 and 2015, which could play a role in access to jobs, along with the average commute times, income levels, and location of jobs in the city (Shearer et. al, 2022).

3 Methodology

3.1 Data

We started our project with commute data collected and published by the Urban Institute (Stern et al., 2020). This data and the subsequent report focuses on four Metropolitan Statistical Areas (MSAs): Seattle, WA, Lansing, MI, Nashville, TN, and Baltimore, MD. These cities are used as case studies to examine transportation access to jobs, specifically among low income workers. They used transportation data from OpenTripPlanner, which creates a model based on transit and road grid data, to calculate travel time between two points, as well as road data from OpenStreetMap, transit data from the Transitland feed registry, and traffic data from INRIX’s 2019 Global Traffic Scorecard. Additionally, they used demographic data from the 2014-18 ACS five year estimates, census defined block group boundaries, and population weighted centroids for each block group from the Missouri Census Data Center. To calculate access to jobs, they included data on opportunities like jobs, hospitals, libraries, and higher education. They include all data collected for each city.

The unequal commute researchers have a nine-step process that they use to calculate the different variables they create. First, they defined the geographic scope, including about a forty-five mile radius to adequately capture workers that could access jobs in the MSAs that they previously defined. Second, they gathered transit and road data as previously discussed to build a representation of the transportation network. Third, they identified start and end points for their routing. Fourth, they calculated travel times, filtering for the fastest itinerary for each route, restricting walk distance within the commute, factoring in possible transfers, and controlling for arrival time. This analysis was set to model travel prior to any effects of the COVID-19 pandemic restrictions. Fifth, they created alternatives for missing routes (like for those who commute from islands into the city of Seattle). Sixth, they adjusted for higher traffic levels during peak commute times. Next, they calculated access to jobs. They summarized the jobs accessible both by public transit and car to low-wage workers by block group, divided by the competitiveness of these jobs or the number of workers who live within a reasonable commuting time of each job. These two equations (by car and transit) were then combined to create an overall job accessibility score for low wage workers by block group. Next, they specifically distinguish between peak and third-shift (late-night) transit accessibility. Finally, they calculated racial disparity in access to jobs by comparing the racial and ethnic distribution of neighborhoods with high levels of spatial mismatch to the racial and ethnic distribution of the whole urban area.

While the Urban Institute’s final dataset included multiple new variables they calculated from the data they gathered, we were particularly interested in one of their variables – spatial mismatch. Spatial mismatch is a normalized measure of the distance between low wage job seekers and access to desirable jobs. Spatial mismatch relies on two factors: how desirable a job is, based on how many people apply for it, and how long it takes to get to the job. Someone who has a high spatial mismatch has a harder time accessing desirable low income jobs, while someone with a low spatial mismatch has easy access to desirable low income jobs. We visualized spatial mismatch in both Seattle and Baltimore, which you can see in Figures 1 and 2.

Figure 1: Spatial Mismatch in the Seattle Area, with 45 mile range

Figure 2: Spatial Mismatch in the Baltimore Area, with 45 mile range

We chose to focus on spatial mismatch from the Urban Institute’s dataset because we were interested in what it might tell us about the two different cities and how they differ in terms of transit and job accessibility. Spatial mismatch is a unique variable that allows for a discussion of multiple factors. It is important to note that spatial mismatch is a normalized measure, and the measure differs between Seattle and Baltimore.

Along with the data on spatial mismatch, we noticed that the Urban Institute included racial information on low income job access. We pulled demographic data from the American Community Survey (ACS) and included percentages of White, Hispanic, Black, Asian, and all other races by block group. The Urban Institute then split the block groups into two categories, with one that included 10% of the block groups with the highest spatial mismatch, and the 90% lower ones falling into the other section. One striking feature we noticed with these categories is that block groups with higher proportions of POC people living in them appeared to have higher spatial mismatches compared to block groups that were white dominant, as you can see in Figure 3. This observation focused our research question, as well as giving us a clear dataset to work with in terms of racial data by block group.

Figure 3: Equity Comparisons, White vs. Nonwhite in Seattle & Baltimore

Figure 3: Equity Comparisons, White vs. Nonwhite in Seattle & Baltimore

We used ACS data from 2014-2018 to match the Urban Institute’s spatial mismatch data, as they used this dataset to for racial demographics and the transportation data to calculate spatial mismatch initially. This survey, conducted by the United States Census Bureau, collects and publishes data over a 5-year period. Their findings are estimates that are published with a 90% confidence interval to account for unreliable and incomplete data. Block groups are the smallest population grouping reported by ACS. This results in some groups with extremely low estimates of people or even no people living in them due to inaccurate data or the block group being located in inhabited areas. In Figure 4, you can see an example of a block group that likely has no residents.

Figure 4: An empty block group in Seattle. Likely no one lives here because it is in the middle of Lake Union

Figure 4: An empty block group in Seattle. Likely no one lives here because it is in the middle of Lake Union

To draw more accurate conclusions with the survey data with our analyses, we decided to remove all block groups that had an estimate of 0 people living in them.

Because the Urban Institute calculated spatial mismatch for both the cities of Baltimore and Seattle as well as the forty-five mile radius surrounding the city, we utilized FIPS codes to determine which counties within or near each city were included. The racial groups we used were estimates of people of a singular race living in a block group. We also pulled the estimated total number of people living in block groups so we could make proportions of all the racial groups. We also made a variable called “POC” that encompasses Black, Asian, Hispanic, and NPI. These are the racial variables we used:

  • White
  • Black
  • Asian
  • Hispanic
  • Native/Pacific Islander (NPI)
  • POC

Additionally, we included educational attainment by block group from the ACS data. We expected block groups with higher levels of educational attainment to have better access to jobs. To encompass as much of the educational attainment data as possible, we also decided to add a variable, “p12”, that encompasses those who only attended 12th grade or lower (and includes people never attended school). We included this variable after the others, but we did not include it in our final models.

These are the educational attainment variables we used:

  • 12th grade or lower, no diploma (p12Prop)
  • 12th grade only, no diploma
  • High school diploma (hsdiplomaProp)
  • Went to a college for a year or less, no degree (college1Prop)
  • Went to a college for 1+ year, no degree (somecollegeProp)
  • GED degree (GEDProp)
  • Associate degree (assProp)
  • Bachelor degree (bachelorProp)

3.2 Testing

We conducted a Welch two sample t-test, a test that is used to determine if two means are different from each other. We assumed the means between the two cities were equal and set our statistical significance level at 0.05. We then began conducting correlation tests to see the relationships between variables and spatial mismatch. We also wanted to see how strong these relationships were and if any of these variables weren’t statistically significant.

The variables interact differently once included with each other, and we moved on to make linear models after performing correlation tests. For Seattle and Baltimore, we used stepwise regression in both directions to fit a model for predicting spatial mismatch. Stepwise regression allows for a program to perform the work of making a linear model, adding and removing predictors based on their significance. All educational variables were included in the stepping process, but to prevent issues of collinearity with racial variables, we tested twice for each city, once using white as the only racial predictor and once using only POC.

The final form of testing that we conducted was Principal Component Analysis, or PCA. PCA reduces the multicollinearity within the variables by extracting the most common characteristics to create new characteristics called components. These components include the information from the original characteristics, but are different combinations of quantitative variables weighted by coefficients based on their contribution to the response variable.

4 Results

4.1 Welch Two Sample t-test

The Welch two sample t-test revealed that the cities in fact had different means, allowing us to reject the null hypothesis that their means were equal, and proceed to conducting further tests to better understand the relationship between the different variables in both Seattle and Baltimore.

4.2 Initial Modeling

Initial correlation testing with spatial mismatch per city confirmed our initial assumptions, primarily that block groups with higher proportions of POC people living in them tended to have higher mismatch, and the correlation tests with individual race groups were significant but poor predictors of spatial mismatch alone. Figures 5 and 6 show the results from these correlation tests.

##                              Lower CI    Upper CI      P-Value
## White                     -0.18845521 -0.11146052 5.755169e-14
## Black                      0.11288392  0.18984517 3.329159e-14
## Asian                     -0.09606204 -0.01754793 4.620049e-03
## Hispanic                   0.14741631  0.22347211 1.165256e-20
## NPI                        0.04005771  0.11833160 7.756715e-05
## POC                        0.09772725  0.17502867 8.707862e-12
## p12                        0.11839016  0.19521916 3.817316e-15
## HS diploma                 0.21346184  0.28728556 8.050373e-37
## HS no diploma              0.04730722  0.12548702 1.614296e-05
## GED                        0.11116761  0.18816914 6.437259e-14
## College < 1 yr.            0.16401161  0.23956811 3.057688e-24
## College > 1 yr. no degree  0.07908440  0.15675611 3.735617e-09
## Associate                  0.07436697  0.15212392 1.511683e-08
## Bachelor                  -0.39233233 -0.32367277 5.275582e-76

Figure 5: Initial Correlation Test for Seattle

##                               Lower CI     Upper CI      P-Value
## White                     -0.024057257  0.064935869 3.674154e-01
## Black                     -0.099333521 -0.010572516 1.531380e-02
## Asian                      0.039503117  0.127908476 2.178973e-04
## Hispanic                   0.004647766  0.093463488 3.044053e-02
## NPI                       -0.077879549  0.011051245 1.405511e-01
## POC                       -0.072611041  0.016348759 2.147415e-01
## p12                       -0.171506141 -0.083930848 1.562294e-08
## HS diploma                -0.045991609  0.043038591 9.480913e-01
## HS no diploma             -0.001602755  0.087263993 5.883872e-02
## GED                       -0.061566110  0.027438310 4.517780e-01
## College < 1 yr.            0.026534053  0.115116966 1.767052e-03
## College > 1 yr. no degree -0.004483653  0.084404293 7.795551e-02
## Associate                  0.103037844  0.190151261 8.096276e-11
## Bachelor                  -0.094607826 -0.005802281 2.675419e-02

Figure 6: Initial Correlation Test for Baltimore

4.3 Linear Modeling: Stepwise Regression

While not resulting in large amount of the variance being explained by the models, we found out that the educational attainment and racial variables were able to explain more of the variation in Seattle (\(R_{ad}^2 = 0.143\) for POC, \(R_{ad}^2 = 0.145\) for white) compared to Baltimore (\(R_{ad}^2 = 0.049\) for POC/white).

The linear models created for Seattle suggested that block groups with higher proportions of white people and people with bachelor degrees would have lower spatial mismatch, which was a fantastic insight in relation to our research question. On the other side, block groups with higher proportions of people who attended a college but got no degree or are POC also had higher spatial mismatch.

## 
## Call:
## lm(formula = spatialmismatch ~ WhiteProp + p12Prop + hsdiplomaProp + 
##     college1Prop + assProp + bachelorProp, data = Seattle_finality)
## 
## Residuals:
##      Min       1Q   Median       3Q      Max 
## -0.09008 -0.02290 -0.00481  0.01688  0.27604 
## 
## Coefficients:
##                Estimate Std. Error t value Pr(>|t|)    
## (Intercept)    0.108030   0.004620  23.382  < 2e-16 ***
## WhiteProp     -0.024493   0.004161  -5.887 4.48e-09 ***
## p12Prop       -0.064151   0.017591  -3.647 0.000271 ***
## hsdiplomaProp  0.026012   0.013008   2.000 0.045640 *  
## college1Prop   0.108310   0.026037   4.160 3.29e-05 ***
## assProp        0.039034   0.020230   1.930 0.053778 .  
## bachelorProp  -0.114293   0.010756 -10.626  < 2e-16 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Residual standard error: 0.03427 on 2470 degrees of freedom
## Multiple R-squared:  0.1467, Adjusted R-squared:  0.1446 
## F-statistic: 70.76 on 6 and 2470 DF,  p-value: < 2.2e-16

Figure 7: Seattle Stepwise Regression Summary

Testing for Baltimore led to the same variables for white and POC. Less of the variability was explained by the model compared to what we saw in Seattle, and only educational variables remained. The coefficient for bachelorProp was still negative, which aligned with Seattle’s models.

## 
## Call:
## lm(formula = spatialmismatch ~ p12Prop + college1Prop + assProp + 
##     bachelorProp, data = Baltimore_finality)
## 
## Residuals:
##       Min        1Q    Median        3Q       Max 
## -0.114608 -0.034311 -0.008945  0.023297  0.237154 
## 
## Coefficients:
##               Estimate Std. Error t value Pr(>|t|)    
## (Intercept)   0.097345   0.004215  23.095  < 2e-16 ***
## p12Prop      -0.133147   0.018594  -7.161 1.14e-12 ***
## college1Prop  0.060337   0.035783   1.686   0.0919 .  
## assProp       0.171452   0.035978   4.765 2.03e-06 ***
## bachelorProp -0.079827   0.014480  -5.513 4.00e-08 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Residual standard error: 0.04615 on 1934 degrees of freedom
## Multiple R-squared:  0.05102,    Adjusted R-squared:  0.04906 
## F-statistic:    26 on 4 and 1934 DF,  p-value: < 2.2e-16

Figure 8: Baltimore Stepwise Regression Summary

After testing for Baltimore, we decided to transform spatial mismatch. Predicting the square root of spatial mismatch made for immensely improved residuals, but they still weren’t entirely normally distributed. We decided to use Principal Component Analysis next to see if we could get better results.

Figure 9: Histogram of Residuals

Figure 9: Histogram of Residuals

Figure 10: Quantile-Quantile Plot of Residuals

Figure 10: Quantile-Quantile Plot of Residuals 

## [1] 1706  440

4.4 Principal Component Analysis

4.4.1 Seattle

4.4.1.1 Creation of Principal Components

When we created our principal components, we noticed that our first component individually explained 41% of the variance and our second component individually explained 32% of the variance. Yet, our third component individually explained 6% of the variance. Therefore, with our first and second components we can account for 74% of the variance in the data.

##          Variables Principal_Component_1 Principal_Component_2
## 1        WhiteProp                  0.29                  0.57
## 2        AsianProp                  0.11                 -0.51
## 3     HispanicProp                 -0.37                 -0.21
## 4        BlackProp                 -0.29                 -0.33
## 5    hsdiplomaProp                 -0.43                  0.20
## 6  hsnodiplomaProp                 -0.26                 -0.06
## 7  somecollegeProp                 -0.21                  0.25
## 8     College1Prop                 -0.26                  0.30
## 9    associateProp                 -0.10                  0.21
## 10    bachelorProp                  0.55                 -0.11

Figure 11: Seattle Principal Component Analysis

As shown in Figure 11, the main variables that contribute to Principal Component 1 are \(\textit{BachelorProp}\), \(\textit{hsdiplomaProp}\), and \(\textit{Hispanic Prop}\). Among these variables, \(\textit{BachelorProp}\) is positive such that \(\textit{BachelorProp} = 0.55\). Conversely, \(\textit{hsdiplomaProp}\), and \(\textit{HispanicProp}\) are all negative such that \(\textit{hsdiplomaProp} = -0.43\), and \(\textit{HispanicProp} = -0.37\).

If there is a higher proportion than average in each block group of low-income job seekers who have attained a bachelor’s degree, then Principal Component 1 will increase. Conversely, if there is a higher proportion in each block group of low-income job seekers than typical who have attained a high school diploma but have no college, then Principal Component 1 will decrease. Additionally, higher proportions of Hispanic low-income job seekers in each block group will also decrease Principal Component 1. Note that \(\textit{BachelorProp}\) and \(\textit{hsdiplomaProp}\) have the highest contrasting weights in Principal Component 1. This implies that Principal Component 1 is measuring some characteristics of the low-income job-seeker’s educational attainment status.

The main variables that contribute to Principal Component 2 are \(\textit{WhiteProp}, \textit{College1Prop}, \textit{AsianProp}\), and \(\textit{BlackProp}\). Both \(\textit{WhiteProp}\) and \(\textit{College1Prop}\) are positive. \(\textit{WhiteProp} = 0.57\) and \(\textit{College1Prop} = 0.30\). So, if there is a higher proportion of low-income job-seekers who are white than is typical across block groups, then Principal Component 2 will increase. Additionally, Principal Component 2 will increase with higher proportions of low-income job-seekers who attended around a year of college but no more. Conversely, \(\textit{AsianProp}\) and \(\textit{BlackProp}\) are both negative, with \(\textit{AsianProp} = -0.51\) and \(\textit{BlackProp} = -0.33\). With higher proportions of Asian low-income job seekers in block groups, Principal Component 2 will decrease, as well as if there are higher proportion of Black low-income job seekers. \(\textit{WhiteProp}\) and \(\textit{AsianProp}\) have the highest contrasting weights in Principal Component 2. We can see that Principal Component 2 is measuring and creating new characteristics regarding low-income job seeker’s race; more specifically White and Asian.

4.4.1.2 Modeling

Keeping both Principal Component 1 and 2 in mind, we created a model to predict spatial mismatch.

First, we created single variable models using Principal Component 1 and Principal Component 2. In these models, Principal Component 2 was not statistically significant, so we moved on to only using Principal Component 1 in the rest of our models.

Using only Principal Component 1, we constructed our next model:

\[\widehat{Spatial Mismatch} = -0.008 \cdot PC1 + 0.078\]

The \(R_{ad}^2 = 0.13\). This \(R_{ad}^2\) was the same as our previous models.

We observed the residuals patterns and noticed curvature towards, so we conducted a polynomial transformation. The interpretability and lower \(R_{ad}^2\) value of our polynomial transformation lead us to a logarithmic transformation. With this transformation, we observe that our residuals are linear and are within our normal distribution of margin of error.

We then decided that in terms of interpretability and higher \(R_{ad}^2\) our logarithmic transformation was the better model for Seattle:

\[\widehat{Spatial Mismatch} = e^{-0.122946 \cdot PC1 -2.663302}\]

Significantly, the coefficient for Principal Component 1 is negative, which implies the variables \(\textit{hsdiplomaProp}\) and \(\textit{HispanicProp }\) have more influence on predicting spatial mismatch.

Figure 12: Predicting Spatial Mismatch in Seattle With Principal Component 1

Figure 12: Predicting Spatial Mismatch in Seattle With Principal Component 1

As shown in Figure 12, because the coefficient for Principal Component 1 is negative, there is a higher spatial mismatch. We see as Principal Component 1 increases, spatial mismatch decreases.

Recall that if there is a higher-than-average proportion in each block group of low-income job seekers who are Hispanic or have high school diplomas, then Principal Component 1 is negative. But, Principal Component 1 is positive if there is a higher than average proportion of low-income job seekers with bachelor degrees in block groups.

So, we observe that low-income job seekers in block groups with higher than typical proportions of Hispanics or high school diplomas holders will tend to have higher spatial mismatch than block groups with higher proportions of low-income job seekers who have bachelor degrees.

4.4.2 Baltimore

4.4.2.1 Creation of Principal Components

With the creation of principal components for Baltimore, we noticed that our first component individually explained 55% of the variance and our second component individually explained 14% of the variance. Yet, our third component individually explained 8% of the variance. Therefore, with our first and second components we can account for 70% of the variance in the data.

##          Variables PrincipalComponent_1 PrincipalComponent_2
## 1        WhiteProp                 0.46                0.430
## 2        AsianProp                 0.23               -0.450
## 3     HispanicProp                 0.01               -0.230
## 4        BlackProp                -0.50               -0.360
## 5    hsdiplomaProp                -0.37                0.320
## 6  hsnodiplomaProp                -0.24               -0.007
## 7  somecollegeProp                -0.25                0.070
## 8     College1Prop                -0.07                0.490
## 9    associateProp                 0.05                0.350
## 10    bachelorProp                 0.46               -0.140

Figure 13: Baltimore Principal Component Analysis

As shown in Figure 13, the main variables that contribute to Principal Component 1 are \(\textit{BachelorProp}, \textit{BlackProp}\), and \(\textit{WhiteProp}\). Amongst these variables \(\textit{BachelorProp}\) and \(\textit{WhiteProp}\) are positive such that \(\textit{BachelorProp} = 0.46\) and \(\textit{WhiteProp} = 0.46\). Conversely, \(\textit{BlackProp}\) is negative such that \(\textit{BlackProp} = -0.50\).

This means that if there is a higher proportion of low-income job seekers per block group than average who have attained a bachelor’s degree, then Principal Component 1 will increase. Additionally, if there is a higher proportion of white low-income job seekers than average per block group, then Principal Component 1 will increase. However, if there is a higher proportion of Black low-income job seekers per block group, then Principal Component 1 will decrease. Our analysis shows that \(\textit{WhiteProp}\) and \(\textit{BlackProp}\) have the highest contrasting weights in Principal Component 1. This implies that Principal Component 1 is measuring some characteristics of the low-income job-seeker’s race and educational status.

The main variables that contribute to Principal Component 2 are \(\textit{WhiteProp}, \textit{College1Prop}\), and \(\textit{AsianProp}\). Out of these variables both \(\textit{WhiteProp}\) and \(\textit{College1Prop}\) are positive. \(\textit{WhiteProp} = 0.43\) and \(\textit{College1Prop} = 0.49\). So, if there is a higher than typical proportion of white low-income job-seekers or low-income job-seekers who attended less than a year of college in each block group, then Principal Component 2 increases. However, \(\textit{AsianProp}\) is negative such that \(\textit{AsianProp} = -0.45\). This means Principal Component 2 will decrease if there is a higher proportion of Asian low-income job seekers than average. Observe that, \(\textit{WhiteProp}\) and \(\textit{AsianProp}\) have the highest contrasting weights in Principal Component 2. We observe that Principal Component 2 is measuring and creating new characteristics regarding low-income job seeker’s race, specifically if they are White or Asian, as well as having some higher education.

4.4.2.2 Modeling

With the observations from Principal Components 1 & 2 in mind, created a model to predict spatial mismatch.

First, we created single variable models using both Principal Components. We noticed that Principal Component 1 was not statistically significant. So, we created a model using only Principal Component 2:

\[\widehat{Spatial Mismatch} = -0.003 \cdot PC2 + 0.086\]

For this model, \(R_{ad}^2 = 0.006\). This is an incredibly low \(R_{ad}^2\), so we added a smoother to see if there were any useful trends in the data.

## `geom_smooth()` using method = 'gam' and formula = 'y ~ s(x, bs = "cs")'
Figure 14: Smoother for Baltimore

Figure 14: Smoother for Baltimore

However, the line in Figure 14 is essentially flat. The only curvature occurring is capturing the few outliers that have a negative Principal Component 2 with lower spatial mismatch. This tells us that Principal Component 2, consisting of proportions of low-income job-seekers who are white, Asian or have less than a year of college, was not significant enough to predict spatial mismatch in Baltimore.

The best \(R_{adj}^2\) we were able to find was \(R_{adj}^2 = 0.01\). This was after a square root transformation on the response variable to align residuals. When we plotted this curve onto our graph, it failed to model the trends in the data.

Therefore, we conclude that race and educational attainment are not good predictors of spatial mismatch in Baltimore.

4.4.3 Seattle and Baltimore

4.4.3.1 Creating the Components

After analyzing Seattle and Baltimore individually, we wanted to investigate if there were any commonalties in variables in predicting spatial mismatch regardless of the city. This lead to our final Principal Component Analysis.

Figure 15: Analyzation of Both Seattle & Baltimore

Figure 15: Analyzation of Both Seattle & Baltimore

In Figure 15, the color scale indicates the impact of contributions to Principal Component 1 and 2. The impacted contribution amongst the different components are \(\textit{WhiteProp}\), \(\textit{BachelorProp}\), and \(\textit{BlackProp}\). These three variables are highly correlated in both cities. This suggests that some characteristics found in these three variables comprise and account for a large portion of the data.

5 Discussion

Recall that our findings in Seattle reveal that race and education affect spatial mismatch When the proportion of White people and those who have bachelor’s degrees are higher, then spatial mismatch is more likely to be lower. Conversely, when proportions of high school graduates, Black, and Hispanic people are higher, spatial mismatch goes up.

These results align with historic oppressive practices. For example, Redlining a discriminatory practice used to deny mortgages, insurance loans and financial services based on certain areas, causes segregation between people of color and white people. Redlining denied people of color access to different neighborhood and forced them to move in different areas of the city. From the 1920’s to the 1940’s, people of color were denied access to Capitol Hill, North Seattle, West Seattle, some parts of South Seattle, Queen Anne, and Madison Park ((Honig, 2021). This resulted in people of color living in restricted areas like the Central District and the International District (Seattle/King County: Mapping Race 1940-2020 - Seattle Civil Rights and Labor History Project, n.d.).

Furthermore, gentrification, a process in which the character of a neighborhood is changed due to the influx of more affluent residents and businesses, has also displaced local residents due to the increased costs of living. This is seen in South Park, a historically predominantly Hispanic neighborhood, where the median home values increased by an average of 47% between 2000 and 2013 (Gentrification and Changing Foodscapes in Seattle, 2020). This disproportionately impacts low-income residents, often primarily people of color, to move farther out of the city into the suburbs.

When the proportion of white people in a block group is higher, it tends to be located in in a more affluent part of a neighborhood, and often means residents have more money to commute to desirable jobs. However, when Black or Hispanic proportions are higher, they tend to be located in less affluent neighborhoods, which results in more barriers to accessing desirable jobs. These conclusions correlate with our findings on spatial mismatch, since desirable low-income jobs are more accessible to white residents, because spatial mismatch is lower. Additionally, having a higher proportion of bachelor’s degrees in a neighborhood is also correlated with spatial mismatch. This again tracks with our assumptions, as white people tend to have greater access to education, especially higher education.

However, in Baltimore, we were unable to find significant results for relationships between the race and education variables and spatial mismatch. This could be due to overall bad commute times in Baltimore, because if all people are having a hard time accessing work, then other factors may not have as large of an influence (Zhang, 2018).

Additionally, Baltimore has a predominantly Black population, which could make it harder to draw relationships between race and education if most people in the area are one race (Yeip, 2015). Baltimore also has a complex history with a low average income compared to other regional cities, population loss, and other challenges (“The Black Butterfly”: Racial Segregation and Investment Patterns in Baltimore, n.d.).

5.1 Limitations

We are unable to compare both Seattle and Baltimore’s relationship and significance because of the different measurements of spatial mismatch in both the cities. We can only see different key factors that are important for both, by observing Principal Components 1 and 2 in the data.

6 References

Balk, G. (2023, May 18). Seattle reclaims title as fastest-growing big city. The Seattle Times. https://www.seattletimes.com/seattle-news/data/seattle-is-once-again-the-fastest-growing-big-city-census-data-shows/

Gentrification and changing foodscapes in Seattle. (2020, July 28). Urban @ UW. https://urban.uw.edu/news/gentrification-and-changing-foodscapes-in-seattle/

Honig, D. (2021, October 29). Redlining in Seattle—HistoryLink.org. HistoryLink.Org. https://www.historylink.org/File/21296

McFadden, D. (2018, December 29). Baltimore trying to stem decades-long disappearing act. AP NEWS. https://apnews.com/article/us-news-race-and-ethnicity-baltimore-neighborhoods-ap-top-news-c852e9751cd744b7a27fa9333c14d1a0

Seattle/King County: Mapping Race 1940-2020—Seattle Civil Rights and Labor History Project. (n.d.). Retrieved June 5, 2023, from https://depts.washington.edu/civilr/maps_race_seattle.htm

Shearer, C., Vey, J. S., & Kim, J. (2022, March 9). Where jobs are concentrating and why it matters to cities and regions. Brookings. https://www.brookings.edu/research/where-jobs-are-concentrating-why-it-matters-to-cities-and-regions/

Stern, A., Stacy, C., Blagg, K., Su, Y., Noble, E., Rainer, M., & Ezike, R. (2020). Access to Opportunity through Equitable Transportation: Technical Appendix. Urban Institute. https://www.urban.org/sites/default/files/2020/10/01/access_to_opportunity_-_technical_appendix.pdf

“The Black Butterfly”: Racial Segregation and Investment Patterns in Baltimore. (n.d.). Urban Institute. Retrieved June 5, 2023, from https://apps.urban.org/features/baltimore-investment-flows/

The Unequal Commute. (2020, October 6). Urban Institute. https://www.urban.org/features/unequal-commute

US Census Bureau. (2023, January 26). American Community Survey Data. Census.gov. https://www.census.gov/programs-surveys/acs/data.html

Yeip, R. (2015, May 1). Baltimore’s Demographic Divide. WSJ. http://graphics.wsj.com/baltimore-demographics/

Zhang, C. (2018). Baltimore commutes again among nation’s longest, Census data show. The Baltimore Sun. https://www.baltimoresun.com/maryland/bs-md-baltimore-commute-20180913-story.html